net/dccp/output.c

   1 /*
   2  *  net/dccp/output.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/config.h>
  14 #include <linux/dccp.h>
  15 #include <linux/kernel.h>
  16 #include <linux/skbuff.h>
  17
  18 #include <net/sock.h>
  19
  20 #include "ackvec.h"
  21 #include "ccid.h"
  22 #include "dccp.h"
  23
  24 static inline void dccp_event_ack_sent(struct sock *sk)
  25 {
  26         inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
  27 }
  28
  29 static inline void dccp_skb_entail(struct sock *sk, struct sk_buff *skb)
  30 {
  31         skb_set_owner_w(skb, sk);
  32         WARN_ON(sk->sk_send_head);
  33         sk->sk_send_head = skb;
  34 }
  35
  36 /*
  37  * All SKB's seen here are completely headerless. It is our
  38  * job to build the DCCP header, and pass the packet down to
  39  * IP so it can do the same plus pass the packet off to the
  40  * device.
  41  */
  42 static int dccp_transmit_skb(struct sock *sk, struct sk_buff *skb)
  43 {
  44         if (likely(skb != NULL)) {
  45                 const struct inet_sock *inet = inet_sk(sk);
  46                 const struct inet_connection_sock *icsk = inet_csk(sk);
  47                 struct dccp_sock *dp = dccp_sk(sk);
  48                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
  49                 struct dccp_hdr *dh;
  50                 /* XXX For now we're using only 48 bits sequence numbers */
  51                 const int dccp_header_size = sizeof(*dh) +
  52                                              sizeof(struct dccp_hdr_ext) +
  53                                           dccp_packet_hdr_len(dcb->dccpd_type);
  54                 int err, set_ack = 1;
  55                 u64 ackno = dp->dccps_gsr;
  56
  57                 dccp_inc_seqno(&dp->dccps_gss);
  58
  59                 switch (dcb->dccpd_type) {
  60                 case DCCP_PKT_DATA:
  61                         set_ack = 0;
  62                         /* fall through */
  63                 case DCCP_PKT_DATAACK:
  64                         break;
  65
  66                 case DCCP_PKT_SYNC:
  67                 case DCCP_PKT_SYNCACK:
  68                         ackno = dcb->dccpd_seq;
  69                         /* fall through */
  70                 default:
  71                         /*
  72                          * Only data packets should come through with skb->sk
  73                          * set.
  74                          */
  75                         WARN_ON(skb->sk);
  76                         skb_set_owner_w(skb, sk);
  77                         break;
  78                 }
  79
  80                 dcb->dccpd_seq = dp->dccps_gss;
  81                 dccp_insert_options(sk, skb);
  82
  83                 skb->h.raw = skb_push(skb, dccp_header_size);
  84                 dh = dccp_hdr(skb);
  85
  86                 /* Build DCCP header and checksum it. */
  87                 memset(dh, 0, dccp_header_size);
  88                 dh->dccph_type  = dcb->dccpd_type;
  89                 dh->dccph_sport = inet->sport;
  90                 dh->dccph_dport = inet->dport;
  91                 dh->dccph_doff  = (dccp_header_size + dcb->dccpd_opt_len) / 4;
  92                 dh->dccph_ccval = dcb->dccpd_ccval;
  93                 /* XXX For now we're using only 48 bits sequence numbers */
  94                 dh->dccph_x     = 1;
  95
  96                 dp->dccps_awh = dp->dccps_gss;
  97                 dccp_hdr_set_seq(dh, dp->dccps_gss);
  98                 if (set_ack)
  99                         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), ackno);
 100
 101                 switch (dcb->dccpd_type) {
 102                 case DCCP_PKT_REQUEST:
 103                         dccp_hdr_request(skb)->dccph_req_service =
 104                                                         dp->dccps_service;
 105                         break;
 106                 case DCCP_PKT_RESET:
 107                         dccp_hdr_reset(skb)->dccph_reset_code =
 108                                                         dcb->dccpd_reset_code;
 109                         break;
 110                 }
 111
 112                 icsk->icsk_af_ops->send_check(sk, skb->len, skb);
 113
 114                 if (set_ack)
 115                         dccp_event_ack_sent(sk);
 116
 117                 DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 118
 119                 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 120                 err = icsk->icsk_af_ops->queue_xmit(skb, 0);
 121                 if (err <= 0)
 122                         return err;
 123
 124                 /* NET_XMIT_CN is special. It does not guarantee,
 125                  * that this packet is lost. It tells that device
 126                  * is about to start to drop packets or already
 127                  * drops some packets of the same priority and
 128                  * invokes us to send less aggressively.
 129                  */
 130                 return err == NET_XMIT_CN ? 0 : err;
 131         }
 132         return -ENOBUFS;
 133 }
 134
 135 unsigned int dccp_sync_mss(struct sock *sk, u32 pmtu)
 136 {
 137         struct dccp_sock *dp = dccp_sk(sk);
 138         /*
 139          * FIXME: we really should be using the af_specific thing to support
 140          *        IPv6.
 141          * mss_now = pmtu - tp->af_specific->net_header_len -
 142          *           sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext);
 143          */
 144         int mss_now = (pmtu - inet_csk(sk)->icsk_af_ops->net_header_len -
 145                        sizeof(struct dccp_hdr) - sizeof(struct dccp_hdr_ext));
 146
 147         /* Now subtract optional transport overhead */
 148         mss_now -= dp->dccps_ext_header_len;
 149
 150         /*
 151          * FIXME: this should come from the CCID infrastructure, where, say,
 152          * TFRC will say it wants TIMESTAMPS, ELAPSED time, etc, for now lets
 153          * put a rough estimate for NDP + TIMESTAMP + TIMESTAMP_ECHO + ELAPSED
 154          * TIME + TFRC_OPT_LOSS_EVENT_RATE + TFRC_OPT_RECEIVE_RATE + padding to
 155          * make it a multiple of 4
 156          */
 157
 158         mss_now -= ((5 + 6 + 10 + 6 + 6 + 6 + 3) / 4) * 4;
 159
 160         /* And store cached results */
 161         dp->dccps_pmtu_cookie = pmtu;
 162         dp->dccps_mss_cache = mss_now;
 163
 164         return mss_now;
 165 }
 166
 167 void dccp_write_space(struct sock *sk)
 168 {
 169         read_lock(&sk->sk_callback_lock);
 170
 171         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
 172                 wake_up_interruptible(sk->sk_sleep);
 173         /* Should agree with poll, otherwise some programs break */
 174         if (sock_writeable(sk))
 175                 sk_wake_async(sk, 2, POLL_OUT);
 176
 177         read_unlock(&sk->sk_callback_lock);
 178 }
 179
 180 /**
 181  * dccp_wait_for_ccid - Wait for ccid to tell us we can send a packet
 182  * @sk: socket to wait for
 183  * @timeo: for how long
 184  */
 185 static int dccp_wait_for_ccid(struct sock *sk, struct sk_buff *skb,
 186                               long *timeo)
 187 {
 188         struct dccp_sock *dp = dccp_sk(sk);
 189         DEFINE_WAIT(wait);
 190         long delay;
 191         int rc;
 192
 193         while (1) {
 194                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 195
 196                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
 197                         goto do_error;
 198                 if (!*timeo)
 199                         goto do_nonblock;
 200                 if (signal_pending(current))
 201                         goto do_interrupted;
 202
 203                 rc = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 204                                             skb->len);
 205                 if (rc <= 0)
 206                         break;
 207                 delay = msecs_to_jiffies(rc);
 208                 if (delay > *timeo || delay < 0)
 209                         goto do_nonblock;
 210
 211                 sk->sk_write_pending++;
 212                 release_sock(sk);
 213                 *timeo -= schedule_timeout(delay);
 214                 lock_sock(sk);
 215                 sk->sk_write_pending--;
 216         }
 217 out:
 218         finish_wait(sk->sk_sleep, &wait);
 219         return rc;
 220
 221 do_error:
 222         rc = -EPIPE;
 223         goto out;
 224 do_nonblock:
 225         rc = -EAGAIN;
 226         goto out;
 227 do_interrupted:
 228         rc = sock_intr_errno(*timeo);
 229         goto out;
 230 }
 231
 232 int dccp_write_xmit(struct sock *sk, struct sk_buff *skb, long *timeo)
 233 {
 234         const struct dccp_sock *dp = dccp_sk(sk);
 235         int err = ccid_hc_tx_send_packet(dp->dccps_hc_tx_ccid, sk, skb,
 236                                          skb->len);
 237
 238         if (err > 0)
 239                 err = dccp_wait_for_ccid(sk, skb, timeo);
 240
 241         if (err == 0) {
 242                 struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 243                 const int len = skb->len;
 244
 245                 if (sk->sk_state == DCCP_PARTOPEN) {
 246                         /* See 8.1.5.  Handshake Completion */
 247                         inet_csk_schedule_ack(sk);
 248                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 249                                                   inet_csk(sk)->icsk_rto,
 250                                                   DCCP_RTO_MAX);
 251                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 252                 } else if (dccp_ack_pending(sk))
 253                         dcb->dccpd_type = DCCP_PKT_DATAACK;
 254                 else
 255                         dcb->dccpd_type = DCCP_PKT_DATA;
 256
 257                 err = dccp_transmit_skb(sk, skb);
 258                 ccid_hc_tx_packet_sent(dp->dccps_hc_tx_ccid, sk, 0, len);
 259         } else
 260                 kfree_skb(skb);
 261
 262         return err;
 263 }
 264
 265 int dccp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 266 {
 267         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk) != 0)
 268                 return -EHOSTUNREACH; /* Routing failure or similar. */
 269
 270         return dccp_transmit_skb(sk, (skb_cloned(skb) ?
 271                                       pskb_copy(skb, GFP_ATOMIC):
 272                                       skb_clone(skb, GFP_ATOMIC)));
 273 }
 274
 275 struct sk_buff *dccp_make_response(struct sock *sk, struct dst_entry *dst,
 276                                    struct request_sock *req)
 277 {
 278         struct dccp_hdr *dh;
 279         struct dccp_request_sock *dreq;
 280         const int dccp_header_size = sizeof(struct dccp_hdr) +
 281                                      sizeof(struct dccp_hdr_ext) +
 282                                      sizeof(struct dccp_hdr_response);
 283         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 284                                                dccp_header_size, 1,
 285                                            GFP_ATOMIC);
 286         if (skb == NULL)
 287                 return NULL;
 288
 289         /* Reserve space for headers. */
 290         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 291
 292         skb->dst = dst_clone(dst);
 293         skb->csum = 0;
 294
 295         dreq = dccp_rsk(req);
 296         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_RESPONSE;
 297         DCCP_SKB_CB(skb)->dccpd_seq  = dreq->dreq_iss;
 298         dccp_insert_options(sk, skb);
 299
 300         skb->h.raw = skb_push(skb, dccp_header_size);
 301
 302         dh = dccp_hdr(skb);
 303         memset(dh, 0, dccp_header_size);
 304
 305         dh->dccph_sport = inet_sk(sk)->sport;
 306         dh->dccph_dport = inet_rsk(req)->rmt_port;
 307         dh->dccph_doff  = (dccp_header_size +
 308                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 309         dh->dccph_type  = DCCP_PKT_RESPONSE;
 310         dh->dccph_x     = 1;
 311         dccp_hdr_set_seq(dh, dreq->dreq_iss);
 312         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dreq->dreq_isr);
 313         dccp_hdr_response(skb)->dccph_resp_service = dreq->dreq_service;
 314
 315         dh->dccph_checksum = dccp_v4_checksum(skb, inet_rsk(req)->loc_addr,
 316                                               inet_rsk(req)->rmt_addr);
 317
 318         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 319         return skb;
 320 }
 321
 322 struct sk_buff *dccp_make_reset(struct sock *sk, struct dst_entry *dst,
 323                                 const enum dccp_reset_codes code)
 324
 325 {
 326         struct dccp_hdr *dh;
 327         struct dccp_sock *dp = dccp_sk(sk);
 328         const int dccp_header_size = sizeof(struct dccp_hdr) +
 329                                      sizeof(struct dccp_hdr_ext) +
 330                                      sizeof(struct dccp_hdr_reset);
 331         struct sk_buff *skb = sock_wmalloc(sk, MAX_HEADER + DCCP_MAX_OPT_LEN +
 332                                                dccp_header_size, 1,
 333                                            GFP_ATOMIC);
 334         if (skb == NULL)
 335                 return NULL;
 336
 337         /* Reserve space for headers. */
 338         skb_reserve(skb, MAX_HEADER + DCCP_MAX_OPT_LEN + dccp_header_size);
 339
 340         skb->dst = dst_clone(dst);
 341         skb->csum = 0;
 342
 343         dccp_inc_seqno(&dp->dccps_gss);
 344
 345         DCCP_SKB_CB(skb)->dccpd_reset_code = code;
 346         DCCP_SKB_CB(skb)->dccpd_type       = DCCP_PKT_RESET;
 347         DCCP_SKB_CB(skb)->dccpd_seq        = dp->dccps_gss;
 348         dccp_insert_options(sk, skb);
 349
 350         skb->h.raw = skb_push(skb, dccp_header_size);
 351
 352         dh = dccp_hdr(skb);
 353         memset(dh, 0, dccp_header_size);
 354
 355         dh->dccph_sport = inet_sk(sk)->sport;
 356         dh->dccph_dport = inet_sk(sk)->dport;
 357         dh->dccph_doff  = (dccp_header_size +
 358                            DCCP_SKB_CB(skb)->dccpd_opt_len) / 4;
 359         dh->dccph_type  = DCCP_PKT_RESET;
 360         dh->dccph_x     = 1;
 361         dccp_hdr_set_seq(dh, dp->dccps_gss);
 362         dccp_hdr_set_ack(dccp_hdr_ack_bits(skb), dp->dccps_gsr);
 363
 364         dccp_hdr_reset(skb)->dccph_reset_code = code;
 365
 366         dh->dccph_checksum = dccp_v4_checksum(skb, inet_sk(sk)->saddr,
 367                                               inet_sk(sk)->daddr);
 368
 369         DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 370         return skb;
 371 }
 372
 373 /*
 374  * Do all connect socket setups that can be done AF independent.
 375  */
 376 static inline void dccp_connect_init(struct sock *sk)
 377 {
 378         struct dst_entry *dst = __sk_dst_get(sk);
 379         struct inet_connection_sock *icsk = inet_csk(sk);
 380
 381         sk->sk_err = 0;
 382         sock_reset_flag(sk, SOCK_DONE);
 383
 384         dccp_sync_mss(sk, dst_mtu(dst));
 385
 386         /*
 387          * FIXME: set dp->{dccps_swh,dccps_swl}, with
 388          * something like dccp_inc_seq
 389          */
 390
 391         icsk->icsk_retransmits = 0;
 392 }
 393
 394 int dccp_connect(struct sock *sk)
 395 {
 396         struct sk_buff *skb;
 397         struct inet_connection_sock *icsk = inet_csk(sk);
 398
 399         dccp_connect_init(sk);
 400
 401         skb = alloc_skb(MAX_DCCP_HEADER + 15, sk->sk_allocation);
 402         if (unlikely(skb == NULL))
 403                 return -ENOBUFS;
 404
 405         /* Reserve space for headers. */
 406         skb_reserve(skb, MAX_DCCP_HEADER);
 407
 408         DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_REQUEST;
 409         skb->csum = 0;
 410
 411         dccp_skb_entail(sk, skb);
 412         dccp_transmit_skb(sk, skb_clone(skb, GFP_KERNEL));
 413         DCCP_INC_STATS(DCCP_MIB_ACTIVEOPENS);
 414
 415         /* Timer for repeating the REQUEST until an answer. */
 416         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 417                                   icsk->icsk_rto, DCCP_RTO_MAX);
 418         return 0;
 419 }
 420
 421 void dccp_send_ack(struct sock *sk)
 422 {
 423         /* If we have been reset, we may not send again. */
 424         if (sk->sk_state != DCCP_CLOSED) {
 425                 struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 426
 427                 if (skb == NULL) {
 428                         inet_csk_schedule_ack(sk);
 429                         inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
 430                         inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
 431                                                   TCP_DELACK_MAX,
 432                                                   DCCP_RTO_MAX);
 433                         return;
 434                 }
 435
 436                 /* Reserve space for headers */
 437                 skb_reserve(skb, MAX_DCCP_HEADER);
 438                 skb->csum = 0;
 439                 DCCP_SKB_CB(skb)->dccpd_type = DCCP_PKT_ACK;
 440                 dccp_transmit_skb(sk, skb);
 441         }
 442 }
 443
 444 EXPORT_SYMBOL_GPL(dccp_send_ack);
 445
 446 void dccp_send_delayed_ack(struct sock *sk)
 447 {
 448         struct inet_connection_sock *icsk = inet_csk(sk);
 449         /*
 450          * FIXME: tune this timer. elapsed time fixes the skew, so no problem
 451          * with using 2s, and active senders also piggyback the ACK into a
 452          * DATAACK packet, so this is really for quiescent senders.
 453          */
 454         unsigned long timeout = jiffies + 2 * HZ;
 455
 456         /* Use new timeout only if there wasn't a older one earlier. */
 457         if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
 458                 /* If delack timer was blocked or is about to expire,
 459                  * send ACK now.
 460                  *
 461                  * FIXME: check the "about to expire" part
 462                  */
 463                 if (icsk->icsk_ack.blocked) {
 464                         dccp_send_ack(sk);
 465                         return;
 466                 }
 467
 468                 if (!time_before(timeout, icsk->icsk_ack.timeout))
 469                         timeout = icsk->icsk_ack.timeout;
 470         }
 471         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 472         icsk->icsk_ack.timeout = timeout;
 473         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 474 }
 475
 476 void dccp_send_sync(struct sock *sk, const u64 seq,
 477                     const enum dccp_pkt_type pkt_type)
 478 {
 479         /*
 480          * We are not putting this on the write queue, so
 481          * dccp_transmit_skb() will set the ownership to this
 482          * sock.
 483          */
 484         struct sk_buff *skb = alloc_skb(MAX_DCCP_HEADER, GFP_ATOMIC);
 485
 486         if (skb == NULL)
 487                 /* FIXME: how to make sure the sync is sent? */
 488                 return;
 489
 490         /* Reserve space for headers and prepare control bits. */
 491         skb_reserve(skb, MAX_DCCP_HEADER);
 492         skb->csum = 0;
 493         DCCP_SKB_CB(skb)->dccpd_type = pkt_type;
 494         DCCP_SKB_CB(skb)->dccpd_seq = seq;
 495
 496         dccp_transmit_skb(sk, skb);
 497 }
 498
 499 /*
 500  * Send a DCCP_PKT_CLOSE/CLOSEREQ. The caller locks the socket for us. This
 501  * cannot be allowed to fail queueing a DCCP_PKT_CLOSE/CLOSEREQ frame under
 502  * any circumstances.
 503  */
 504 void dccp_send_close(struct sock *sk, const int active)
 505 {
 506         struct dccp_sock *dp = dccp_sk(sk);
 507         struct sk_buff *skb;
 508         const gfp_t prio = active ? GFP_KERNEL : GFP_ATOMIC;
 509
 510         skb = alloc_skb(sk->sk_prot->max_header, prio);
 511         if (skb == NULL)
 512                 return;
 513
 514         /* Reserve space for headers and prepare control bits. */
 515         skb_reserve(skb, sk->sk_prot->max_header);
 516         skb->csum = 0;
 517         DCCP_SKB_CB(skb)->dccpd_type = dp->dccps_role == DCCP_ROLE_CLIENT ?
 518                                         DCCP_PKT_CLOSE : DCCP_PKT_CLOSEREQ;
 519
 520         if (active) {
 521                 dccp_skb_entail(sk, skb);
 522                 dccp_transmit_skb(sk, skb_clone(skb, prio));
 523         } else
 524                 dccp_transmit_skb(sk, skb);
 525 }